diagnostics

Article An Easy and Reliable Strategy for Making Type I Signature Analysis Comparable among Research Centers

Alessia Pin 1 , Lorenzo Monasta 2 , Andrea Taddio 1,3, Elisa Piscianz 4 , Alberto Tommasini 3,* and Alessandra Tesser 4

1 Department of Medicine, Surgery and Health Sciences, University of Trieste, 34127 Trieste, Italy 2 Clinical Epidemiology and Public Health Research Unit, Institute for Maternal and Child Health—IRCCS “Burlo Garofolo”, 34137 Trieste, Italy 3 Department of Paediatrics, Institute for Maternal and Child Health—IRCCS “Burlo Garofolo”, 34137 Trieste, Italy 4 Department of Advanced Diagnostic and Clinical Trials, Institute for Maternal and Child Health—IRCCS “Burlo Garofolo”, 34137 Trieste, Italy * Correspondence: [email protected]

 Received: 9 August 2019; Accepted: 3 September 2019; Published: 4 September 2019 

Abstract: Interferon-stimulated genes (ISGs) are a set of genes whose transcription is induced by interferon (IFN). The measure of the expression of ISGs enables calculating an IFN score, which gives an indirect estimate of the exposition of cells to IFN-mediated inflammation. The measure of the IFN score is proposed for the screening of monogenic interferonopathies, like the Aicardi-Goutières syndrome, or to stratify subjects with systemic lupus erythematosus to receive IFN-targeted treatments. Apart from these scenarios, there is no agreement on the diagnostic value of the score in distinguishing IFN-related disorders from diseases dominated by other types of . Since the IFN score is currently measured in several research hospitals, merging experiences could help define the potential of scoring IFN inflammation in clinical practice. However, the IFN score calculated at different laboratories may be hardly comparable due to the distinct sets of IFN-stimulated genes assessed and to different controls used for data normalization. We developed a reliable approach to minimize the inter-laboratory variability, thereby providing shared strategies for the IFN signature analysis and allowing different centers to compare data and merge their experiences.

Keywords: interferon signature score; inter-laboratory variability; data sharing; systemic lupus erythematosus; interferonopathies; biostatistics

1. Introduction Type I interferon (IFN) production is part of the innate immune response to viruses or intracellular bacteria, which is triggered by the sensing of pathogen-associated nucleic acids [1]. Even though the identification of IFNs dates back to the 50s–60s [2], the description of a group of mendelian disorders with dysregulated IFN-mediated inflammation has only recently shed on the fine regulation of the production and action of these cytokines [3]. Of note, this new group of disorders, known as type I interferonopathies, displays significant phenotypic overlaps with both systemic lupus erythematosus (SLE) and congenital viral infections of the TORCH (Toxoplasmosis, Rubella, Cytomegalovirus, Herpes simplex) and HIV (human immunodeficiency virus) groups [3,4]. Type I interferonopathies are marked by the hyper-expression of a set of genes (IFN-stimulated genes, ISGs) in inflamed tissue and often in peripheral blood, leading to the definition of the so-called

Diagnostics 2019, 9, 113; doi:10.3390/diagnostics9030113 www.mdpi.com/journal/diagnostics Diagnostics 2019, 9, 113 2 of 14

“IFN signature” [5,6]. The IFN signature was firstly defined in subjects with SLE to assess the level of IFN related inflammation and to help stratify patients to receive IFN targeted treatments [7–9]. Since then, the measure of expression of the ISGs (IFN signature analysis) is increasingly used in biomedical research centers, as well as for the functional classification of other conditions characterized by a type I IFN dysregulation [10], to distinguish such conditions from classical inflammatory disorders predominantly mediated by other cytokines, like Tumor Factor α and 1 (i.e., inflammatory bowel diseases, rheumatoid arthritis, and periodic ) [11]. Different ISG sets were identified to evaluate interferon-mediated autoinflammation and are frequently restricted to 5-6 targeted genes [6,10,12,13], suitable, for example, for the discrimination of Aicardi-Goutières syndrome (AGS) [6,10]. However, IFN signature analysis could be extended to larger gene lists [7] or even restricted to just one single gene, particularly when directly assessed in affected tissues, as in the case of dermatomyiositis [14–16]. Even though the IFN signature measure has become widely available at research hospitals, there is no consensus for the selection of calibration controls. Thus, it is hard to compare data among distinct centers and to estimate the potential of IFN signature testing for discriminating among inflammatory conditions in the clinical practice. For example, thousands of subjects with antiphospholipid syndrome have been described in multicenter studies [17], while the interferon score has been separately studied in several small series without allowing the merging of results [18–22]. The main problem hindering the use of IFN signature analysis for in vitro diagnostics is the expression of data relativized to independent healthy control(s) in each laboratory, leading to unpredictable inter-laboratory variability. It is logical to assume that the use of pooled cDNA from healthy donors can represent a convenient strategy for calibrating Real Time quantitative PCR (qPCR) for the assessment of the IFN score. However, it can be difficult to predict the optimal number of samples to be pooled, which requires minimizing variability between one pool and another prepared at distinct laboratories, as already pointed out by others [23,24]. This study aims to investigate, through laboratory, bioinformatic, and statistical analyses, a reliable approach to minimize the variability that can be observed in inter-laboratory assays. The final goal is providing shared recommendations for IFN signature analysis and interpretation of data in the clinical practice, thereby allowing data sharing among reference centers and improving knowledge on IFN-related disorders.

2. Materials and Methods The study is part of the IRCCS Burlo Garofolo project RC #24/2017, approved by the Institutional Review Board and by the Friuli Venezia Giulia Independent Ethical Committee (2018-SPER-079-BURLO, N. 0039851, approved on 12 December 2018). All investigations were performed after obtaining written informed consent from volunteers and patients or their parents/guardians.

2.1. Subjects On wet IFN signature analysis was assessed by quantitative PCR (qPCR) in ten young-aged healthy subjects (Dataset A, ten out of eleven individuals, five males and five females). To establish whether data from qPCR and RNAseq analysis were comparable, IFN signature on wet (by qPCR) and in silico (by RNAseq) analysis was performed in twenty subjects with inflammatory diseases, such as systemic lupus erythematosus (SLE), interferonopathies or inflammatory bowel diseases (IBD), and patients’ relatives, recruited at our center (Dataset E). A brief description of patients’ clinical diagnosis is displayed in Table S1. To increase healthy donors numerosity, in silico IFN signature investigation (by RNAseq) has been performed in twenty healthy individuals, collected at our center Dataset A (four out of eleven individuals), and selected from different whole blood RNA-sequencing (RNAseq) open-access web-based datasets: we sorted another three datasets, while considering exclusively healthy control samples, accessible at ArrayExpress (accession number E-MTAB-5735, Dataset B, five individuals) and Diagnostics 2019, 9, 113 3 of 14 at the Gene Expression Omnibus (GEO) (accession number GSE112057, Dataset C, nine individuals; GSE90081, Dataset D, two individuals). Specification about gender was not available for all the samples. However, sex has been easily inferred by expression analysis of the sex-specific genes RPS4Y1 and USP9Y. The dataset composition is shown in Table1.

Table 1. Dataset composition: subjects, size, methods and purposes.

Datasets Subjects n Total (F/M) Method (n) Purpose To test the variability of expression qPCR (10) A—Data of from our of the six ISGs in a healthy donor Healthy donors 11 (5/6) RNAseq (3) center; Accession: # small group available at out center To increase the healthy donor group RNAseq (1) size, to improve the power of the variability measurement To increase the healthy donor group B—Accession: Healthy donors 5 (2/3) RNAseq (5) size, to improve the power of the E-MTAB-5735 variability measurement To increase the healthy donor group C—Accession: Healthy donors 12 (6/6) RNAseq (9) size, to improve the power of the GSE112057 variability measurement To increase the healthy donor group D—Accession: Healthy donors 12 (12/0) RNAseq (2) size, to improve the power of the GSE90081 variability measurement E—Patients and To compare IFN signature results qPCR (20) patient’s relatives Patients 20 (9/11) between qPCR and RNAseq RNAseq (20) recruited at our center analyses # data not present in open-access web-based datasets.

2.2. Sample Collection, RNA Isolation and cDNA Preparation Peripheral blood was collected in PAXgene Blood RNA Tubes (PreAnalytiX, Hombrechtikon, Switzerland) and, after two-hours incubation at room temperature, tubes were frozen at 20 C until − ◦ processing. Total RNA was extracted with PAXgene Blood RNA Kit (PreAnalytiX, Switzerland), following the manufacturer’s instructions, and quantified with NanoDrop Spectrophotometer (Thermo Fisher, Waltham, MA, USA). RNA integrity was checked using an Agilent Technologies 2100 Bioanalyzer. Up to 1 µg of total RNA was retro-transcribed using SensiFAST cDNA Synthesis Kit (Bioline, London, UK).

2.3. IFN Signature Analysis The expression of six IFN-stimulated genes was assessed by qPCR using AB 7500 Real Time PCR System (Applied Biosystems, Waltham, MA, USA), TaqMan Gene Expression Master Mix (Applied Biosystems, USA) and UPL Probes (Roche, Basel, Switzerland) for IFI27, IFI44L, IFIT1, ISG15, RSAD2, and SIGLEC1. Using AB 7500 Real Time PCR software, each target quantity was normalized with the expression level of HPRT1 and G6PD, and the relative quantification (RQ) was conducted relating to a ∆∆Ct “calibrator” sample (mix of ten healthy controls, Dataset A) using the 2− method [25]. The median fold change of the six genes was used to calculate the IFN score for each patient.

2.4. RNAseq Analysis Transcriptome sequencing was performed using the TruSeq Stranded mRNA Sample Preparation (Illumina, San Diego, CA, USA) and sequenced on a NovaSeq 6000 platform (Illumina, San Diego, CA, USA), generating 2X100 bp paired-end reads (30 million reads per sample) in twenty subjects from Dataset E (patients and patients’ relatives) and four out of eleven controls from Dataset A. Diagnostics 2019, 9, 113 4 of 14

RNAseq raw data (either our data and open-access web-based data) workflow was conducted as follows: quality control by FastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/), quality filtering by Trim Galore (https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/), read alignment to hg38 using annotation from GENECODE v.31 (https://www.gencodegenes.org/) with STAR [26], reads counting into genes by featureCounts [27]. Data of patients with autoinflammatory diseases and three healthy individuals of our dataset were normalized and analyzed for differentially expressed genes by DESeq2 [28]. From the result table, we only considered the ISGs and evaluated their relative fold changes on each patient compared to the set of controls. To assess the ISGs expression variability within the group of twenty healthy subjects, shortlisted from the datasets described above, we determined the expression values for each gene, normalized by Fragments Per Kilobase per Million mapped reads (FPKM) method with edgeR (rpkm function) [29,30], using the values from the “Length” column, in the featureCounts’ output, for the calculation. Principal component analysis (PCA), useful for data visualization, was conducted with DESeq2, to define the overall variability between samples.

2.5. Statistical Analyses Considering that each of the six genes measured was expressed on a different scale, we decided to calculate the sample size based on the coefficient of variation, instead than the mean and the standard deviation. We further hypothesized that different runs did not increase the variation in comparing the samples, assuming the only origin of variability to be represented by the subjects’ heterogeneity. To determine the statistical power for data obtained by qPCR and RNAseq, we computed the noncentrality parameter (λ) using GPower 3.1.9.2. software [31,32], with a generic two-tailed t-test, given α = 0.05, β = 0.2, and degrees of freedom equal N-1. If the noncentrality parameter under these conditions (reference value, “λref”) resulted in being lower than the one calculated on our samples, we considered the sample size as appropriate. GraphPad Prism 6 software was employed for χ2 contingency analysis; p-values <0.05 were considered significant. To identify the appropriate sample size for variability assessment, we computed λ for increasing numerosity (up to forty) using GPower 3.1.9.2. (Heinrich-Heine University Düsseldorf, Germany), and determined a “plateau value” by an exponential decay function (GraphPad Prism 6 software, La Jolla California USA).

3. Results

3.1. Variability Assessment in IFN-Stimulated Genes Expression in Healthy Controls (Dataset A) Analyzed by qPCR The variability of expression of the six ISGs (IFI27, IFI44L, IFIT1, ISG15, RSAD2, SIGLEC1) was assessed in ten healthy controls processed by on wet qPCR analysis. Five out of six genes showed low variability coefficients and noncentrality parameters (λ) that fulfilled the analysis criteria (as described in Materials and Methods, Section 2.5) (Table2). Only IFI44L did not comply with the analysis parameters, presenting higher variability and a lower λ than the reference value (λref) (Figure1) Diagnostics 2019, 9, 113 5 of 14 Diagnostics 2019, 9, x FOR PEER REVIEW 5 of 14

Table 2. Variability assessment for interferon-stimulated genes (ISGs) expression values quantified in Table 2. Variability assessment for interferon-stimulated genes (ISGs) expression values quantified in ten out of eleven healthy subjects from Dataset A by qPCR. ten out of eleven healthy subjects from Dataset A by qPCR.

IFI27IFI27 IFI44LIFI44L IFIT1 IFIT1 ISG15 ISG15 RSAD2 RSAD2 SIGLEC1 SIGLEC1 MeanMean 4.57 4.57 0.720.72 2.18 2.18 4.61 4.61 1.24 1.24 5.14 5.14 SDSD 1.04 1.04 0.910.91 1.05 1.05 0.60 0.60 1.20 1.20 0.42 0.42 VariabilityVariability coe coefficientfficient 0.23 0.23 1.26 1.26 0.48 0.48 0.13 0.13 0.96 0.96 0.08 0.08 λ (λn (=n 10)= 10) 13.92 13.92 2.522.52 6.566.56 24.47 24.47 3.28 3.28 38.99 38.99 λref: 3.15 λref: 3.15 SD: standard deviation; λref: noncentrality parameter (λ) calculated based on α = 0.05, β = 0.2, and SD: standard deviation; λref: noncentrality parameter (λ) calculated based on α = 0.05, β = 0.2, and degrees of freedomdegrees equalof freedom N-1. Sample equal size N-1. is consideredSample size as appropriateis considered when as λappropriatecomputed on when each geneλ computed is higher thanon eachλref. Thegene value is higher below thanλref isλ highlightedref. The value in bold. below λref is highlighted in bold.

Figure 1. Graphical representation of the noncentrality parameter (λ) calculated for each ISG by on wetFigure qPCR 1. Graphical analysis. Therepresentationλref for ten subjectsof the noncentrality is displayed byparameter the dashed (λ) linecalculated and reported for each in ISG the figure.by on Samplewet qPCR size analysis. is considered The λref as for appropriate ten subjects when is displayedλ computed by the on dashed each gene line is and higher reported than λ inref. the figure. Sample size is considered as appropriate when λ computed on each gene is higher than λref. Thus, ten healthy controls could not be considered an appropriate sample size to represent an idealThus, healthy ten population, healthy controls in which could the not physiologically be considered floating an appropriate expression sample values size of the to ISGsrepresent present an acceptableideal healthy variability. population, For in this which reason, the we physiologically should increase floating the numerosity expression of values healthy of controls the ISGs to present obtain aacceptable suitable pool variability. in which For the this gene reason, expression we should variability increase is the minimized. numerosity of healthy controls to obtain a suitable pool in which the gene expression variability is minimized. 3.2. IFN-Stimulated Genes Expression Evaluated by qPCR or RNAseq Analysis Are Comparable 3.2. IFN-Stimulated Genes Expression Evaluated by qPCR or RNAseq Analysis Are Comparable To improve the power of the variability measurement, we decided to take advantage of RNAseq open-accessTo improve web-based the power data, of as the an variability easy source measurem to increaseent, the we number decided of to healthy take advantage subjects toof calculateRNAseq ISGsopen-access interindividual web-based di data,fferences. as an easy This source choice to came increase from the comparisons number of healthy between subjects the relative to calculate ISGs foldISGs changeinterindividual assessed differences. in the same This twenty choice subjects came (Datasetfrom comparisons E) by both between qPCR and the RNAseqrelative ISGs analysis, fold selectingchange assessed the same in set the of same three outtwenty of eleven subjects healthy (Dataset controls E) by from both Dataset qPCR A, and to normalizeRNAseq analysis, data for bothselecting techniques. the same set of three out of eleven healthy controls from Dataset A, to normalize data for both Sometechniques. subjects showed different relative expression values for the same gene calculated by on wet qPCRSome and insubjects silico RNAseq,showed different but the overall relative results expressi of theon IFNvalues signatures for the same (IFN scores)gene calculated were extremely by on consistentwet qPCR between and in thesilico two RNAseq, techniques but for the each overall individual results (χ of2 contingency the IFN signatures analysis p(IFN-value scores)= 0.405, were not significant).extremely consistent The comparability between the of IFNtwo scorestechniques is easily for explained, each individual considering (χ2 contingency that these valuesanalysis represent p-value the= 0.405, median not ofsignificant). the six relative The ISGscomparability quantifications, of IFN and scores they is broadly easily reflectexplained, the overexpression considering that status these in thevalues analyzed represent sample the (Tablemedian3). Thus,of the thesix tworelative methods ISGs provided quantification the sames, and trend they in genebroadly expression reflect the in subjectsoverexpression presenting status low, in intermediate, the analyzed and sample high IFN(Table signatures, 3). Thus, as the indicated two methods by the threeprovided representative the same graphstrend in in gene Figure expression2 (Subject n.9:in subjects low IFN presenting signature; Subjectlow, intermediate, n.14: intermediate and high IFN IFN signature; signatures, Subject as n.11:indicated high by IFN the signature). three representative graphs in Figure 2 (Subject n.9: low IFN signature; Subject n.14: intermediate IFN signature; Subject n.11: high IFN signature).

Diagnostics 2019, 9, 113 6 of 14

Diagnostics 2019, 9, x FOR PEER REVIEW 6 of 14 Table 3. Comparison of interferon (IFN) scores determined in twenty subjects by both qPCR and RNAseqTable 3. byComparisonχ2 contingency of interferon analysis (IFN) (p-value scores= 0.405, determin not significant).ed in twenty subjects by both qPCR and RNAseq by χ2 contingency analysis (p-value = 0.405, not significant). IFN Score Subject n. IFN Score Subject n. In Silico RNAseq On Wet qPCR In Silico RNAseq On Wet qPCR 1 3.26 5.01 1 3.26 5.01 2 6.79 10.05 2 6.79 10.05 3 7.12 9.85 43 8.587.12 10.749.85 54 1.558.58 1.3710.74 65 1.111.55 0.971.37 76 0.551.11 0.670.97 87 0.220.55 0.190.67 98 1.140.22 1.050.19 109 3.681.14 2.601.05 1110 77.733.68 82.612.60 1211 44.0377.73 94.2582.61 1312 17.4444.03 16.4894.25 14 15.44 16.35 13 17.44 16.48 15 37.43 49.98 14 15.44 16.35 16 37.44 84.99 1715 1.0737.43 1.1849.98 1816 2.3937.44 4.7084.99 1917 3.011.07 1.711.18 2018 1.822.39 1.824.70 19 3.01 1.71 Mean 19.83 13.59 SD20 31.191.82 20.331.82 VariabilityMean 19.83 13.59 1.57 1.50 coefficientSD 31.19 20.33 Variability coefficientSD: standard deviation. 1.57 1.50 SD: standard deviation.

Figure 2. Representative display of low (a), intermediate (b) and high (c) IFN signatures analyzed by bothFigure qPCR 2. Representative and RNAseq display for each of subject. low (a), For intermediate the optimal (b graphical) and high representation (c) IFN signatures of all analyzed histograms, by theboth scales qPCR of and values RNAseq are set for di eachfferent. subject. The For IFN the scores optimal computed graphical for representation each subject are of reportedall histograms, in the legendthe scales of theof values figures. are set different. The IFN scores computed for each subject are reported in the legend of the figures. 3.3. Preliminary Analysis for Sample Selection and Variability Assessment in IFN-Stimulated Genes Expression Analyzed3.3. Preliminary by RNAseq Analysis for Sample Selection and Variability Assessment in IFN-Stimulated Genes ExpressionGiven Analyzed the comparability by RNAseq of data between qPCR and RNAseq, we exploited the availability of largeGiven open-access the comparability web repositories of data containing between RNAseq qPCR and data RNAseq, to calculate we the exploited proper samplethe availability size for the of assessmentlarge open-access of the IFNwebscore repositories with an containing acceptable RNAseq inter-laboratory data to variability.calculate the proper sample size for the assessmentTo select the of appropriatethe IFN score sample with an size, acceptable we firstly inter-laboratory calculated the noncentralityvariability. parameter (λ) on a numerosityTo select up the to appropriate forty individuals sample using size, GPowerwe firstly software. calculated Then, the noncentrality we determine parameter the plateau (λ value) on a ofnumerosity the exponential up to forty decay individuals function of using the previously GPower software. computed Then,λ values. we determine The provided the plateau plateau value (3.01) of correspondsthe exponential to λdecayfor sample function size of nthe= previously15 (Figure 3computed), leading λ us values. to consider The provided fifteen subjects plateau as(3.01) an appropriatecorresponds sampleto λ for size. sample For moresize n experimental = 15 (Figure strength, 3), leading we us considered to consider both fifteenn = 15 subjects and n = as20 an in appropriate sample size. For more experimental strength, we considered both n = 15 and n = 20 in the following analyses. We did not further increase the sample size over twenty subjects, whereas exceeding this number might bring difficulties in term of donors’ collection.

Diagnostics 2019, 9, 113 7 of 14

the following analyses. We did not further increase the sample size over twenty subjects, whereas Diagnosticsexceeding 2019, 9, x FOR this numberPEER REVIEW might bring difficulties in term of donors’ collection. 7 of 14

λ Figure 3.Figure Plateau 3. Plateau value value(reported (reported as asa red a red dot) dot) thresh thresholdold (reported (reported as aas dotted a dotted vertical vertical line) of line) of λ computed for sample sizes up to forty subjects. Each sample size is represented by a black dot. In blue computedthe for exponential sample decaysizes functionup to forty curve subjects. of the previously Each sample computed sizeλ .is represented by a black dot. In blue the exponential decay function curve of the previously computed λ. We performed a principal component analysis (PCA), a data visualization analysis, to evaluate We theperformed total expression a principal variation component of the six ISGs analysis and to define (PCA), the most a data homogeneous visualization set of twentyanalysis, healthy to evaluate individuals. This investigation allows the detection of possible rare outliers with the highest variance the total expression variation of the six ISGs and to define the most homogeneous set of twenty that might not be considered as suitable controls to study IFN signature. healthy individuals.RNAseq recordsThis investigation have been chosen allows considering the detection the presence of possible of similar rare features outliers such with as blood the highest variancecollection that might type, not RNA be considered extraction protocol as suitable and library controls selection, to study to reduce IFN assignature. much as possible the RNAseqtechnical records procedure have variability been chosen (Table3). considering the presence of similar features such as blood collection type,As a RNA first attempt, extraction we investigated protocol alland the library RNAseq selection, samples of to healthy reduce donors as much from Dataset as possible A the (data from our center, n = 4/11), Dataset B (E-MTAB-5735, n = 5) and Dataset C (GSE112057, n = 12), technical procedure variability (Table 3). twenty-one specimens in total. Figure4a displays the PCA results showing the overall ISGs expression As variabilitya first attempt, between we individuals. investigated The analysis all the exhibited RNAseq a highersamples variance of healthy in three outdonors of twenty-one from Dataset A (data fromsubjects: our center, we thus n decided = 4/11), not Dataset to include B (E-MTAB-5735, these three samples n in= further5) and studies,Dataset collecting C (GSE112057, eighteen n = 12), twenty-onesamples specimens that were in suitable total. for Figure our purpose. 4a displays To get the the proper PCA numerosity results showing (n = 20), we the examine overall ISGs expressionDataset variability D (GSE90081, betweenn = 12)indi andviduals. we ran The the sameanalysis analysis exhibited again, obtaininga higher avariance satisfactory in levelthree out of of variation, without outliers, between datasets and among individuals. From these preliminary twenty-one subjects: we thus decided not to include these three samples in further studies, collecting observations, we randomly chose two out of twelve samples from Dataset D, combining them with eighteendata samples previously that selected. were suitable Again, we for observed our purpose. an acceptable To geneget the expression proper variability numerosity among (n our = 20), we examinefinal Dataset twenty-controls-sized D (GSE90081, groupn = 12) (Table and4), we as shown ran the in Figuresame4 b.analysis again, obtaining a satisfactory level of variation, without outliers, between datasets and among individuals. From these preliminary observations, we randomly chose two out of twelve samples from Dataset D, combining them with data previously selected. Again, we observed an acceptable gene expression variability among our final twenty-controls-sized group (Table 4), as shown in Figure 4b.

DiagnosticsDiagnostics 20192019,, 99,, 113x FOR PEER REVIEW 88 of of 14

Figure 4. Overall ISGs expression variance between individuals. PCA of the first selection (a), with Figure 4. Overall ISGs expression variance between individuals. PCA of the first selection (a), with evidence of the three outliers from Dataset C, and of the final twenty-controls-sized group (b), showing evidence of the three outliers from Dataset C, and of the final twenty-controls-sized group (b), twenty samples uniformly distributed (4 from Dataset A, 5 from Dataset B, 9 from Dataset C and 2 showing twenty samples uniformly distributed (4 from Dataset A, 5 from Dataset B, 9 from Dataset from Dataset D). C and 2 from Dataset D). Table 4. Detailed report of final twenty-controls-sized healthy subject groups. Table 4. Detailed report of final twenty-controls-sized healthy subject groups. Datasets Subjects RNAseq Details Datasets Subjects Whole Blood RNAseq Details Authors & Female Male RNAseq library collectionWhole Blood/RNA Read Length AccessionAuthors & (n = 9)Female (n =Male11) preparationRNAseq /libraryplatform Read collection/RNAextraction Accession (n = 9) (n = 11) preparation/platform Length PAXgeneextraction blood A—Data from Illumina TruSeq stranded RNA Paired-end our center; 2 2 PAXgene blood mRNA library A—Data from tube/PAXgene Illumina TruSeq Paired-end100 bp reads Accession:# RNA protocol/Novaseq our center; 2 2 Blood RNA Kit stranded mRNA library 100 bp tube/PAXgene B—RoderoAccession:# MP, PAXgene blood protocol/Novaseq reads Illumina TruSeq stranded et al., 2017; BloodRNA RNA Kit Paired-end 2 3 mRNA library Accession:B—Rodero MP, PAXgenetube/PAXgene blood Illumina TruSeq 75 bp reads protocol/Illumina HiSeq 2000 E-MTAB-5735et al., 2017; BloodRNA RNA Kit stranded mRNA library Paired-end 2 3 C—MoAccession: A., et E- tube/PAXgeneTempus protocol/IlluminaIllumina TruSeq stranded HiSeq 75 bp reads al., 2018; Tube/Tempus mRNA library Paired-end MTAB-5735 3 6 Blood RNA Kit 2000 Accession: SpinTempus isolation protocolIllumina/Illumina TruSeq HiSeq Rapid 100 bp reads GSE112057C—Mo A., et al., RNA kit Run Paired-end Tube/Tempus stranded mRNA library 2018; Accession: 3 6 100 bp D—Shchetynsky PAXgeneSpin isolation blood protocol/IlluminaStandard illumina TruSeq HiSeq K., et al.,GSE112057 2017; RNA RNA protocol, following readsPaired-end 2 - Accession: tubeRNA/PAXgene kit PolyA enrichmentRapid Run/Illumina 100 bp reads GSE90081 Blood miRNA kit StandardHiSeq 2000illumina D—Shchetynsky PAXgene blood TruSeq RNA protocol, Paired-end K., et al., 2017; # data not present in open-accessRNA web-based datasets. 2 - following PolyA 100 bp Accession: tube/PAXgene enrichment/Illumina reads We calculatedGSE90081 the λref for larger samplesBlood of miRNA healthy kit controls (fifteen and twenty subjects) using HiSeq 2000 GPower software, and the variability of the ISGs expression in fifteen and twenty healthy controls # data not present in open-access web-based datasets. processed by in silico RNAseq analysis. Table5 shows that all the variability coe fficients are considerably low andWe allcalculatedλ calculated the λ fulfilledref for larger the analysis samples parameters of healthy (Figure controls5) (expression (fifteen and values twenty of singlesubjects) gene using for GPower software, and the variability of the ISGs expression in fifteen and twenty healthy controls

Diagnostics 2019, 9, x FOR PEER REVIEW 9 of 14 processed by in silico RNAseq analysis. Table 5 shows that all the variability coefficients are considerably low and all λ calculated fulfilled the analysis parameters (Figure 5) (expression values ofDiagnostics single 2019gene, 9 ,for 113 each control are displayed in Table S2. Thus, we can hypothesize that pooling9 of 14 together samples from fifteen or twenty healthy subjects could also be considered a proper sample size in qPCR analyses. each control are displayed in Table S2. Thus, we can hypothesize that pooling together samples from fifteenTable or twenty 5. Variability healthy assessment subjects for could IFN-stimulated also be considered genes expression a proper values sample evaluated size in qPCRin fifteen analyses. and twenty healthy subjects by in silico RNAseq analysis. Table 5. Variability assessment for IFN-stimulated genes expression values evaluated in fifteen and

twenty healthy subjects by in silicoIFI27 RNAseq IFI44L analysis. IFIT1 ISG15 RSAD2 SIGLEC1 Mean 0.35 1.14 5.06 30.17 2.21 1.17 SD 0.23 IFI270.28 IFI44L 1.29 IFIT1 12.44 ISG15 RSAD20.86 SIGLEC10.55 Variability n = 15 Mean0.64 0.350.25 1.14 0.26 5.06 0.41 30.17 2.210.39 0.47 1.17 coefficientSD 0.23 0.28 1.29 12.44 0.86 0.55 n = 15 Variabilityλ coefficient 6.09 0.6415.64 0.25 15.16 0.26 9.39 0.41 0.399.93 8.21 0.47 Λref: 3.01λ 6.09 15.6415.16 9.39 9.93 8.21 Mean 0.33 1.26 5.17 30.09 2.47 1.22 Λref: 3.01 SD 0.22 0.38 1.38 14.41 1.00 0.50 VariabilityMean 0.33 1.26 5.17 30.09 2.47 1.22 n = 20 0.67 0.30 0.27 0.48 0.41 0.41 coefficientSD 0.22 0.38 1.38 14.41 1.00 0.50 n = 20 Variability coefficient 0.67 0.30 0.27 0.48 0.41 0.41 λ 6.68 14.70 16.77 9.34 10.98 10.90 λref: 2.95λ 6.68 14.7016.77 9.34 10.98 10.90 SD: standard deviation;λref: 2.95 λref: noncentrality parameter calculated based on α = 0.05, β = 0.2, and degreesSD: standard of freedom deviation; equalλref: noncentralityN-1. Sample parameter size is considered calculated based as appropriate on α = 0.05, βwhen= 0.2, λ and computed degrees of on freedom each geneequal is N-1. higher Sample than size λref. is considered as appropriate when λ computed on each gene is higher than λref.

Figure 5. Graphical representation of λ calculated for each ISG by in silico RNAseq analysis. The λrefs Figure 5. Graphical representation of λ calculated for each ISG by in silico RNAseq analysis. The λrefs for fifteen and twenty subjects are displayed by the red dashed line and reported in the figure. Sample for fifteen and twenty subjects are displayed by the red dashed line and reported in the figure. Sample size is considered appropriate when λ computed on each gene is higher than λref. size is considered appropriate when λ computed on each gene is higher than λref. 3.4. Pooling Twenty Subjects Could Be Considered an Optimal Strategy to Minimize Gene Expression 3.4.Variability Pooling among Twenty Healthy Subjects Controls Could forBe Considered on Wet IFN an Signature Optimal AnalysisStrategy to Minimize Gene Expression Variability among Healthy Controls for on Wet IFN Signature Analysis We checked whether the values of mean, SD and variability coefficient obtained by on wet qPCR on tenWe controls checked met whether the λ criteria the values considering of mean,n SD= 15 and and variabilityn = 20 subjects coefficient as already obtained calculated, by on wet assuming qPCR onthat ten the controls variability met the coe ffiλ criteriacient does considering not change n = as 15 the and sample n = 20 subjects size increases as already (Table calculated,6). assuming that the variability coefficient does not change as the sample size increases (Table 6).

Diagnostics 2019, 9, 113 10 of 14

Diagnostics 2019, 9, x FOR PEER REVIEW 10 of 14 Table 6. Estimated variability assessment for ISGs expression values in fifteen and twenty healthy Tablesubjects 6. forEstimated qPCR analysis variability assuming assessment the same for variabilityISGs expression coeffi cientvalues for in increasing fifteen and sample twenty size. healthy subjects for qPCR analysis assuming the same variability coefficient for increasing sample size. IFI27 IFI44L IFIT1 ISG15 RSAD2 SIGLEC1 IFI27 IFI44L IFIT1 ISG15 RSAD2 SIGLEC1 Mean 4.57 0.72 2.18 4.61 1.24 5.14 Mean 4.57 0.72 2.18 4.61 1.24 5.14 SD 1.04 0.91 1.05 0.60 1.20 0.42 Variability coeSDffi cient 0.231.04 1.260.91 0.481.05 0.130.60 1.20 0.96 0.42 0.08 Variability coefficient 0.23 1.26 0.48 0.13 0.96 0.08 λ (n = 15) 17.05 3.08 8.04 29.97 4.02 47.75 λ (n = 15) 17.05 3.08 8.04 29.97 4.02 47.75 λref:λ 3.01ref: 3.01 λ (n λ= (20)n = 20) 19.6919.69 3.563.56 9.289.28 34.6134.61 4.64 4.64 55.14 55.14 λref: 2.95 λref: 2.95 Mean, standard deviation (SD) and variability coefficient values previously determined on ten Mean, standard deviation (SD) and variability coefficient values previously determined on ten subjects are reported subjectsin the table. areλ reportedref: noncentrality in the table. parameter λref: (λ noncentrality) calculated based parameter on α = 0.05, (λ)β calculated= 0.2, and degreesbased on of freedomα = 0.05, equal β = 0.2,N-1. and Sample degrees size is of considered freedom appropriateequal N-1. when Sampλlecomputed size is considered on each gene appropriate is higher than whenλref. λ computed on each gene is higher than λref. The analysis analysis provides provides quite quite good good results results for for both both sample sample sizes sizes tested, tested, even even if IFI44L if IFI44L showedshowed a λ λ λ (3.08)a (3.08) very very close close to λ toref ref(3.01). (3.01). For For this this reason, reason, we we can can cons considerider more more adequate adequate to to increase increase the numerosity to twenty subjects, a st stillill easy-to-gather number of healthy donors. Thus, we could could assert assert thatthat twenty twenty healthy healthy controls controls could could be be considered a a suitable sample size for IFN signature analysis performed by qPCR, as predicted by in silico RNAseq (Figure6 6).).

Figure 6. Graphical representation of λ calculated for each ISG by in silico RNAseq analysis and Figurehypothesized 6. Graphical qPCR representation assessment considering of λ calculated fifteen (fora) andeach twenty ISG by (b in) subjects. silico RNAseq The λrefs analysis values and are hypothesizeddisplayed by theqPCR red assessment dashed line considering and reported fifteen in the (a) figure.and twenty Sample (b) sizesubjects. is considered The λrefs appropriate values are displayedwhen λ computed by the red on dashed each gene line is and higher reported than λ ref.in the figure. Sample size is considered appropriate when λ computed on each gene is higher than λref. 4. Discussion 4. DiscussionThe clinical employment of the IFN signature is strictly related to the screening of pathological conditionsThe clinical characterized employment by typeof the I IFN signature dysregulation is strictly [24]. related However, to the several screening studies of pathological have been conditionscarried out characterized to associate the by IFN-related type I IFN inflammation dysregulation with [24]. specific However, clinical several or laboratory studies have features been of carriedrheumatologic out to associate conditions, the likeIFN-related systemic inflammation lupus erythematosus, with specific primary clinical antiphospholipid or laboratory features syndrome, of rheumatologicSjögren syndrome, conditions, rheumatoid like systemic arthritis, lupus autoimmune erythematosus, myositis primary and systemic antiphospholipid sclerosis [18 syndrome,,20,33–38]. SjögrenSome Authors syndrome, proposed rheumatoid that the arthritis, assessment autoimmune of IFN inflammation myositis and maysystemic help sclerosis identify [18,20,33–38]. subgroups of Somepatients Authors with a betterproposed response that tothe specific assessment treatments, of IFN as Binflammation cell targeted therapiesmay help [ 39identify–42]. Moreover, subgroups since of patientsmost anti-inflammatory with a better response or immunomodulatory to specific treatments, agents as have only targeted a weak therapies effect on [39–42]. IFN inflammation, Moreover, sincethe calculation most anti-inflammatory of the IFN score mayor immunomodulatory also serve to guide targeted agents therapyhave only approaches a weak with effect novel on drugs IFN inflammation,like the inhibitorscalculation [10 of]. the IFN score may also serve to guide targeted therapy approaches with However,novel drugs there like is Janus no consensus Kinase inhibitors on a shared [10]. and validated method to classify different inflammatory conditionsHowever, by transcriptome there is no analysis.consensus Crow on anda shared collaborators and validated used pooled method cDNA to from classify healthy different donors inflammatoryas calibrator forconditions qPCR, andby transcriptome after assessing analys a largeis. Crow number and ofcollaborator healthy controlss used pooled and patients cDNA from with healthyAicardi-Gouti donorsères as Syndromecalibrator calculatedfor qPCR, aand fixed after cut-o assessingff value ofa large normality number suitable of healthy for the controls screening and of patientsinterferonopathies with Aicardi-Goutières [6]. However, Syndrome this cut-off calculatedhas been validated a fixed cut-off to facilitate value the of detectionnormality of suitable monogenic for the screening of interferonopathies [6]. However, this cut-off has been validated to facilitate the detection of monogenic interferonopathies and not to assess IFN inflammation in other conditions.

Diagnostics 2019, 9, 113 11 of 14 interferonopathies and not to assess IFN inflammation in other conditions. Moreover, the reference to locally pooled control cDNA may make it difficult to compare results obtained in distinct centers. Even though multiple ISG panels have been described either in peripheral blood cells or affected tissues, many laboratories set their assays by analyzing a minimal set of 5-6 targeted-genes, usually including the set proposed by Crow et al. This set has been applied to thousands of analyses and its potential for screening of monogenic interferonopathies is established. After the normalization of results on at least two housekeeping genes, the main source of variability limiting interlaboratory comparison of data consists in the use of different controls for data normalization. Indeed, this is a remarkable problem, considering that the potential role of IFN signature analysis in clinical practice can be defined only by sharing data among research centers, comparing or merging case series. Of course, the best option for the future should rely on the development of industrially manufactured kits validated for In Vitro Diagnostics (IVD) and usable worldwide with the same reference values, not only in the most advanced research areas. Conversely, only the analysis and comparison of data available at various centers can tell industries whether the development of such diagnostic kits is worthwhile or not. Thus, we focused on a strategy that could be immediately applied in biomedical laboratories to facilitate the sharing of experience and minimize the inter-laboratory variability. We investigated if using a pool of biological samples from healthy controls could solve the data normalization issue, which is the main source of inter-laboratory variability. We proposed that this strategy could “equalize” the differences in gene expression that physiologically occur among individuals. For this purpose, we evaluated if any pool of healthy controls more than a given number n of subjects could be suitable to level differences, through an approach based on laboratory (qPCR), bioinformatic (RNAseq), and statistical data integration analyses. Starting from the ISGs expression assessment in peripheral blood from ten healthy subjects by on wet qPCR analysis, we found that this sample size is not suitable for equalizing the variable expression levels of all the six ISGs in healthy volunteers. To find the appropriate number of samples to be pooled together with low enough variability, we further investigated if public data from RNAseq can be exploited to expand our analysis. We thus compared relative ISGs fold change values calculated by qPCR and RNAseq analysis for the same subjects with the same calibrator, showing that the two methods yield comparable results with very low variability between each other. Previous studies compared gene expression measurements generated by in silico RNAseq and on wet qPCR assays, showing consistent results between the two methods for most genes, with the only exception of some genes generally characterized by small size and low expression levels. None of the ISGs is included in the list of genes with inconsistent estimation of expression according to the Authors [43]. Our results confirmed that RNAseq and qPCR generate consistent results in the assessment of ISGs expression levels. Thus, based on the literature and on our preliminary results, we considered RNAseq as a good asset to increase the number of healthy subjects on which to calculate interindividual differences in ISGs expression, exploiting both RNAseq samples available at our center and open-access web-based data. The results of our study support the choice of pooling twenty healthy controls for the normalization of the assay, allowing to express results as relative to a “standard set of controls”. Of note, we validated this sample size only for the selected set of six ISGs in peripheral blood cells. The expression of other genes included in larger panels, may present higher variability among donors and between different affected tissues. However, the same procedure that we have described can also be used to define the optimal sample size for other transcription profiles, for interferonopathies or for other rheumatologic disorders. The performance of the proposed twenty-controls-sized standard could be improved by analyzing all the donors separately before pooling, and by performing cluster variance analysis, which can enable excluding rare donors with outlier variance.

Supplementary Materials: The following are available online at http://www.mdpi.com/2075-4418/9/3/113/s1. Diagnostics 2019, 9, 113 12 of 14

Author Contributions: Conceptualization: A.T. (Alberto Tommasini), A.P. (Alessia Pin), A.T. (Alessandra Tesser); methodology: A.P. (Alessia Pin), A.T. (Alessandra Tesser), L.M.; software: A.P. (Alessia Pin), L.M.; validation: A.P. (Alessia Pin), A.T. (Alessandra Tesser), E.P.; resources: A.T. (Alberto Tommasini); data curation: A.P. (Alessia Pin), A.T. (Alessandra Tesser), L.M., E.P.; writing—original draft preparation: A.P. (Alessia Pin), A.T. (Alessandra Tesser), L.M.; writing—review and editing: A.T. (Alberto Tommasini), E.P., A.T. (Andrea Taddio); supervision: A.T. (Alberto Tommasini), A.T. (Andrea Taddio), E.P.; project administration: A.T. (Alberto Tommasini), A.T. (Andrea Taddio), A.T. (Alessandra Tesser); funding acquisition: A.T. (Alberto Tommasini), A.T. (Andrea Taddio). Funding: This research was funded by the Institute for Maternal and Child Health IRCCS “Burlo Garofolo” RC #24/2017. Acknowledgments: We thank Stefano Volpi from IRCCS Istituto Giannina Gaslini (Genova, Italy) for the long-term support and scientific confrontation and Remo Sanges from International School for Advanced Studies (SISSA, Trieste, Italy) for the bioinformatic analysis support. Conflicts of Interest: The authors declare no conflict of interest.

References

1. Stetson, D.B.; Medzhitov, R. Type I in host defense. Immunity 2006, 25, 373–381. [CrossRef] [PubMed] 2. Isaacs, A.; Lindenmann, J. Virus interference. I. The interferon. Proc. R. Soc. Lond. B Biol. Sci. 1957, 147, 258–267. [PubMed] 3. Crow, Y.J. Type I interferonopathies: A novel set of inborn errors of immunity. Ann. N. Y. Acad. Sci. 2011, 1238, 91–98. [CrossRef] 4. Crow, Y.J.; Black, D.N.; Ali, M.; Bond, J.; Jackson, A.P.; Lefson, M.; Michaud, J.; Roberts, E.; Stephenson, J.B.; Woods, C.G.; et al. Cree encephalitis is allelic with Aicardi-Goutiéres syndrome: Implications for the pathogenesis of disorders of interferon α metabolism. J. Med. Genet. 2003, 40, 183–187. [CrossRef][PubMed] 5. Baechler, E.C.; Batliwalla, F.M.; Karypis, G.; Gaffney, P.M.; Ortmann, W.A.; Espe, K.J.; Shark, K.B.; Grande, W.J.; Hughes, K.M.; Kapur, V.; et al. Interferon-inducible gene expression signature in peripheral blood cells of patients with severe lupus. Proc. Natl. Acad. Sci. USA 2003, 100, 2610–2615. [CrossRef][PubMed] 6. Rice, G.I.; Forte, G.M.; Szynkiewicz, M.; Chase, D.S.; Aeby, A.; Abdel-Hamid, M.S.; Ackroyd, S.; Allcock, R.; Bailey, K.M.; Balottin, U.; et al. Assessment of interferon-related biomarkers in Aicardi-Goutières syndrome associated with mutations in TREX1, RNASEH2A, RNASEH2B, RNASEH2C, SAMHD1, and ADAR: A case-control study. Lancet Neurol. 2013, 12, 1159–1169. [CrossRef] 7. Yao, Y.; Higgs, B.W.; Morehouse, C.; de Los Reyes, M.; Trigona, W.; Brohawn, P.; White, W.; Zhang, J.; White, B.; Coyle, A.J.; et al. Development of Potential Pharmacodynamic and Diagnostic Markers for Anti-IFN-α Monoclonal Antibody Trials in Systemic Lupus Erythematosus. Hum. Genom. Proteom. 2009, 2009.[CrossRef][PubMed] 8. Furie, R.; Khamashta, M.; Merrill, J.T.; Werth, V.P.; Kalunian, K.; Brohawn, P.; Illei, G.G.; Drappa, J.; Wang, L.; Yoo, S.; et al. , an Anti-Interferon-α Receptor Monoclonal Antibody, in Moderate-to-Severe Systemic Lupus Erythematosus. Arthritis Rheumatol. 2017, 69, 376–386. [CrossRef] 9. Merrill, J.T.; Furie, R.; Werth, V.P.; Khamashta, M.; Drappa, J.; Wang, L.; Illei, G.; Tummala, R. Anifrolumab effects on rash and arthritis: Impact of the type I interferon gene signature in the phase IIb MUSE study in patients with systemic lupus erythematosus. Lupus Sci. Med. 2018, 5, e000284. [CrossRef] 10. Rice, G.I.; Melki, I.; Frémond, M.L.; Briggs, T.A.; Rodero, M.P.; Kitabayashi, N.; Oojageer, A.; Bader-Meunier, B.; Belot, A.; Bodemer, C.; et al. Assessment of Type I Interferon Signaling in Pediatric Inflammatory Disease. J. Clin. Immunol. 2017, 37, 123–132. [CrossRef] 11. Kim, H.; Sanchez, G.A.; Goldbach-Mansky, R. Insights from Mendelian Interferonopathies: Comparison of CANDLE, SAVI with AGS, Monogenic Lupus. J. Mol. Med. 2016, 94, 1111–1127. [CrossRef][PubMed] 12. Feng, X.; Wu, H.; Grossman, J.M.; Hanvivadhanakul, P.; FitzGerald, J.D.; Park, G.S.; Dong, X.; Chen, W.; Kim, M.H.; Weng, H.H.; et al. Association of increased interferon-inducible gene expression with disease activity and lupus nephritis in patients with systemic lupus erythematosus. Arthritis Rheum. 2006, 54, 2951–2962. [CrossRef][PubMed] Diagnostics 2019, 9, 113 13 of 14

13. Higgs, B.W.; Liu, Z.; White, B.; Zhu, W.; White, W.I.; Morehouse, C.; Brohawn, P.; Kiener, P.A.; Richman, L.; Fiorentino, D.; et al. Patients with systemic lupus erythematosus, myositis, rheumatoid arthritis and share activation of a common type I interferon pathway. Ann. Rheum. Dis. 2011, 70, 2029–2036. [CrossRef][PubMed] 14. Greenberg, S.A.; Pinkus, J.L.; Pinkus, G.S.; Burleson, T.; Sanoudou, D.; Tawil, R.; Barohn, R.J.; Saperstein, D.S.; Briemberg, H.R.; Ericsson, M.; et al. Interferon-α/β-mediated innate immune mechanisms in dermatomyositis. Ann. Neurol. 2005, 57, 664–678. [CrossRef][PubMed] 15. Salajegheh, M.; Kong, S.W.; Pinkus, J.L.; Walsh, R.J.; Liao, A.; Nazareno, R.; Amato, A.A.; Krastins, B.; Morehouse, C.; Higgs, B.W.; et al. Interferon-stimulated gene 15 (ISG15) conjugates proteins in dermatomyositis muscle with perifascicular atrophy. Ann. Neurol. 2010, 67, 53–63. [CrossRef] 16. Uruha, A.; Nishikawa, A.; Tsuburaya, R.S.; Hamanaka, K.; Kuwana, M.; Watanabe, Y.; Suzuki, S.; Suzuki, N.; Nishino, I. Sarcoplasmic MxA expression: A valuable marker of dermatomyositis. Neurology 2017, 88, 493–500. [CrossRef][PubMed] 17. Cervera, R.; Serrano, R.; Pons-Estel, G.J.; Ceberio-Hualde, L.; Shoenfeld, Y.; de Ramón, E.; Buonaiuto, V.; Jacobsen, S.; Zeher, M.M.; Tarr, T.; et al. Morbidity and mortality in the antiphospholipid syndrome during a 10-year period: A multicentre prospective study of 1000 patients. Ann. Rheum. Dis. 2015, 74, 1011–1018. [CrossRef][PubMed] 18. Palli, E.; Kravvariti, E.; Tektonidou, M.G. Type I Interferon Signature in Primary Antiphospholipid Syndrome: Clinical and Laboratory Associations. Front. Immunol. 2019, 10, 487. [CrossRef] 19. Bernales, I.; Fullaondo, A.; Marín-Vidalled, M.J.; Ucar, E.; Martínez-Taboada, V.; López-Hoyos, M.; Zubiaga, A.M. Innate immune response gene expression profiles characterize primary antiphospholipid syndrome. Genes Immun. 2008, 9, 38–46. [CrossRef] 20. Grenn, R.C.; Yalavarthi, S.; Gandhi, A.A.; Kazzaz, N.M.; Núñez-Álvarez, C.; Hernández-Ramírez, D.; Cabral, A.R.; McCune, W.J.; Bockenstedt, P.L.; Knight, J.S. Endothelial progenitor dysfunction associates with a type I interferon signature in primary antiphospholipid syndrome. Ann. Rheum. Dis. 2017, 76, 450–457. [CrossRef] 21. van den Hoogen, L.L.; Fritsch-Stork, R.D.; Versnel, M.A.; Derksen, R.H.; van Roon, J.A.; Radstake, T.R. Monocyte type I interferon signature in antiphospholipid syndrome is related to proinflammatory monocyte subsets, hydroxychloroquine and statin use. Ann. Rheum. Dis. 2016, 75, e81. [CrossRef][PubMed] 22. Knight, J.S.; Meng, H.; Coit, P.; Yalavarthi, S.; Sule, G.; Gandhi, A.A.; Grenn, R.C.; Mazza, L.F.; Ali, R.A.; Renauer, P.; et al. Activated signature of antiphospholipid syndrome neutrophils reveals potential therapeutic target. JCI Insight 2017, 2.[CrossRef][PubMed] 23. Fryer, J.F.; Baylis, S.A.; Gottlieb, A.L.; Ferguson, M.; Vincini, G.A.; Bevan, V.M.; Carman, W.F.; Minor, P.D. Development of working reference materials for clinical virology. J. Clin. Virol. 2008, 43, 367–371. [CrossRef] [PubMed] 24. Kim, H.; de Jesus, A.A.; Brooks, S.R.; Liu, Y.; Huang, Y.; VanTries, R.; Montealegre Sanchez, G.A.; Rotman, Y.; Gadina, M.; Goldbach-Mansky, R. Development of a Validated Interferon Score Using NanoString Technology. J. Interferon Res. 2018, 38, 171–185. [CrossRef][PubMed] 25. Livak, K.J.; Schmittgen, T.D. Analysis of relative gene expression data using real-time quantitative PCR and the 2(-Delta Delta C(T)) Method. Methods 2001, 25, 402–408. [CrossRef][PubMed] 26. Dobin, A.; Davis, C.A.; Schlesinger, F.; Drenkow, J.; Zaleski, C.; Jha, S.; Batut, P.; Chaisson, M.; Gingeras, T.R. STAR: Ultrafast universal RNA-seq aligner. Bioinformatics 2013, 29, 15–21. [CrossRef][PubMed] 27. Liao, Y.; Smyth, G.K.; Shi, W. featureCounts: An efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 2014, 30, 923–930. [CrossRef] 28. Love, M.I.; Huber, W.; Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 2014, 15, 550. [CrossRef] 29. Robinson, M.D.; McCarthy, D.J.; Smyth, G.K. edgeR: A Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics 2010, 26, 139–140. [CrossRef] 30. McCarthy, D.J.; Chen, Y.; Smyth, G.K. Differential expression analysis of multifactor RNA-Seq experiments with respect to biological variation. Nucleic Acids Res. 2012, 40, 4288–4297. [CrossRef] 31. Faul, F.; Erdfelder, E.; Lang, A.G.; Buchner, A. G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behav. Res. Methods 2007, 39, 175–191. [CrossRef][PubMed] Diagnostics 2019, 9, 113 14 of 14

32. Faul, F.; Erdfelder, E.; Buchner, A.; Lang, A.G. Statistical power analyses using G*Power 3.1: Tests for correlation and regression analyses. Behav. Res. Methods 2009, 41, 1149–1160. [CrossRef][PubMed] 33. Bodewes, I.L.A.; Björk, A.; Versnel, M.A.; Wahren-Herlenius, M. Innate immunity and interferons in the pathogenesis of Sjögren’s syndrome. Rheumatology 2019.[CrossRef][PubMed] 34. Nezos, A.; Gravani, F.; Tassidou, A.; Kapsogeorgou, E.K.; Voulgarelis, M.; Koutsilieris, M.; Crow, M.K.; Mavragani, C.P. Type I and II interferon signatures in Sjogren’s syndrome pathogenesis: Contributions in distinct clinical phenotypes and Sjogren’s related lymphomagenesis. J. Autoimmun. 2015, 63, 47–58. [CrossRef][PubMed] 35. Cantaert, T.; van Baarsen, L.G.; Wijbrandts, C.A.; Thurlings, R.M.; van de Sande, M.G.; Bos, C.; van der Pouw Kraan, T.C.; van der Pouw, T.K.; Verweij, C.L.; Tak, P.P.; et al. Type I interferons have no major influence on humoral autoimmunity in rheumatoid arthritis. Rheumatology 2010, 49, 156–166. [CrossRef][PubMed] 36. Castañeda-Delgado, J.E.; Bastián-Hernandez, Y.; Macias-Segura, N.; Santiago-Algarra, D.; Castillo-Ortiz, J.D.; Alemán-Navarro, A.L.; Martínez-Tejada, P.; Enciso-Moreno, L.; Garcia-De Lira, Y.; Olguín-Calderón, D.; et al. Type I Interferon Gene Response Is Increased in Early and Established Rheumatoid Arthritis and Correlates with Autoantibody Production. Front. Immunol. 2017, 8, 285. [CrossRef] 37. Ekholm, L.; Vosslamber, S.; Tjärnlund, A.; de Jong, T.D.; Betteridge, Z.; McHugh, N.; Plestilova, L.; Klein, M.; Padyukov, L.; Voskuyl, A.E.; et al. Autoantibody Specificities and Type I Interferon Pathway Activation in Idiopathic Inflammatory Myopathies. Scand. J. Immunol. 2016, 84, 100–109. [CrossRef][PubMed] 38. Brkic, Z.; van Bon, L.; Cossu, M.; van Helden-Meeuwsen, C.G.; Vonk, M.C.; Knaapen, H.; van den Berg, W.; Dalm, V.A.; Van Daele, P.L.; Severino, A.; et al. The signature is present in systemic sclerosis before overt fibrosis and might contribute to its pathogenesis through high BAFF gene expression and high collagen synthesis. Ann. Rheum. Dis. 2016, 75, 1567–1573. [CrossRef][PubMed] 39. Quartuccio, L.; Mavragani, C.P.; Nezos, A.; Gandolfo, S.; Tzioufas, A.G.; De Vita, S. Type I interferon signature may influence the effect of on immunoglobulin levels, including rheumatoid factor in Sjögren’s syndrome. Clin. Exp. Rheumatol. 2017, 35, 719–720. 40. de Jong, T.D.; Vosslamber, S.; Blits, M.; Wolbink, G.; Nurmohamed, M.T.; van der Laken, C.J.; Jansen, G.; Voskuyl, A.E.; Verweij, C.L. Effect of prednisone on type I interferon signature in rheumatoid arthritis: Consequences for response prediction to rituximab. Arthritis Res. Ther. 2015, 17, 78. [CrossRef] 41. Wright, H.L.; Thomas, H.B.; Moots, R.J.; Edwards, S.W. Interferon gene expression signature in rheumatoid arthritis neutrophils correlates with a good response to TNFi therapy. Rheumatology 2015, 54, 188–193. [CrossRef][PubMed] 42. Raterman, H.G.; Vosslamber, S.; de Ridder, S.; Nurmohamed, M.T.; Lems, W.F.; Boers, M.; van de Wiel, M.; Dijkmans, B.A.; Verweij,C.L.; Voskuyl,A.E. The interferon type I signature towards prediction of non-response to rituximab in rheumatoid arthritis patients. Arthritis Res. Ther. 2012, 14, R95. [CrossRef][PubMed] 43. Everaert, C.; Luypaert, M.; Maag, J.L.V.; Cheng, Q.X.; Dinger, M.E.; Hellemans, J.; Mestdagh, P. Benchmarking of RNA-sequencing analysis workflows using whole-transcriptome RT-qPCR expression data. Sci. Rep. 2017, 7, 1559. [CrossRef][PubMed]

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).